HeroApply
Auto Apply to hundreds of jobs in a few hours
English (US)EspañolItalianoPortuguêsFrançaisDeutsch日本語한국어
Sign up

Atmospheric Scientist interview questions and how to answer them

An atmospheric science interview feels a lot like a thesis defense with a hiring decision attached. Expect people to push on your physics, poke at your code and ask why you trusted a result. Here's what each round checks and how to answer the questions that come up again and again.

11 questions7 minute read

Part 1

1.Walk us through your thesis or most recent project, starting from the question.

Why they ask

It's the fastest way to see whether you drove the work or followed instructions, and it gives them material for every follow-up.

How to answer

  • Open with the scientific question in one plain sentence, before any method
  • Explain why you chose that model, dataset or instrument over the obvious alternative
  • Name the result and how confident you are in it
  • Say what you'd do differently now, and mean it

2.Here's this morning's sounding on a Skew-T. Will we get thunderstorms this afternoon?

Why they ask

Reading a sounding is basic literacy in the field, and it shows whether you can turn physics into a judgment call.

How to answer

  • Talk through moisture, the lapse rate and any capping inversion out loud
  • Estimate instability and say what surface heating would need to reach to break the cap
  • Mention shear and what kind of storms it would support
  • Give a clear yes, no or conditional answer, and name what would change your mind

3.What's the difference between a parameterized process and a resolved one, and when has that choice changed your results?

Why they ask

Every model result depends on schemes for clouds, convection, radiation and turbulence. They want to know you understand where the model is guessing.

How to answer

  • Define it simply: processes smaller than the grid get represented by a scheme
  • Give a real example, like switching microphysics or boundary layer schemes in WRF
  • Describe how the output shifted and how you decided which run to trust
  • Connect it to grid spacing, including the gray zone where convection is partly resolved
4.Explain data assimilation to someone who's never heard of it.

Why they ask

It checks your understanding and your ability to explain it to a manager, client or funder in the same breath.

How to answer

  • Use an everyday analogy: blending a best guess with fresh but imperfect measurements
  • Explain that each side gets weighted by how much you trust it
  • Name a method you know, like ensemble Kalman filtering or variational assimilation, without drowning in math
  • Say why it matters: a better starting state makes a better forecast
5.How would you check whether a model bias you found is real and not an artifact of how you verified it?

Why they ask

Verification mistakes are common and expensive. They want to see scientific suspicion aimed at your own work.

How to answer

  • Check the observations first: station siting, instrument changes and quality control flags
  • Compare like with like, matching grid cells to point observations carefully and handling time zones
  • Test against an independent dataset or a different verification metric
  • Look at whether the bias holds across seasons, regimes and locations

Part 2

6.A wind farm client says your forecast cost them money yesterday. What do you do?

Why they ask

Private-sector roles live on client trust. They want to hear accountability without panic or defensiveness.

How to answer

  • Listen first and find out exactly which hours and which numbers went wrong
  • Pull the forecast, the observations and the model guidance you had at the time
  • Explain honestly what was knowable then and what wasn't
  • Offer a concrete fix, like a better ramp alert or a probabilistic product, if the review supports one
7.How do you make a model run or an analysis reproducible months later?

Why they ask

Papers, audits and permit reviews all depend on someone rerunning your work. Plenty of groups have been burned by runs nobody could recreate.

How to answer

  • Keep code and namelists in Git, with the commit recorded for every run
  • Pin software environments with conda or containers
  • Store the input data version and the exact command that produced each output
  • Write a short README a stranger could follow
8.Your job on the cluster dies partway through a long simulation. Walk us through what you do.

Why they ask

HPC work is a daily reality, and they want someone who can debug without waiting for help.

How to answer

  • Read the scheduler logs and the model's own error output first
  • Check the usual suspects: wall-time limits, memory, disk quota and a numerical instability like a CFL violation
  • Restart from the last restart file rather than from scratch
  • Change one thing at a time and write down what fixed it
9.How would you explain the uncertainty in a regional climate projection to a city planner?

Why they ask

More and more atmospheric work lands in front of engineers and officials who have to make decisions from it.

How to answer

  • Separate the sources: scenario choice, model differences and natural variability
  • Show a range, not a single line, and say which end matters for their decision
  • Explain what downscaling adds and what it can't fix
  • Translate into their terms, like design storms or cooling demand
10.Tell us about a time your analysis turned out to be wrong.

Why they ask

Everyone in this field has been wrong. They want to know if you catch it, own it and fix the process.

How to answer

  • Pick a real mistake with real stakes, like a units error or a misaligned time axis
  • Explain how it was caught and by whom
  • Describe what you told the people who had used the result
  • Name the check you added so it wouldn't happen again

Part 3

11.Why this group, and not a university or a different sector?

Why they ask

Research groups, agencies and companies all worry about hiring someone who'll leave once a better fit shows up.

How to answer

  • Name something specific the group does, like its field campaigns, its operational model or its client base
  • Connect it to a thread running through your own work
  • Be honest about the trade-off you're choosing, such as less publishing or more client contact

Questions to ask them

Ask at least two. It shows you're picking them too.

  • How is this position funded, and how long does that money run?
  • What computing resources will I have, and how long do jobs wait in the queue?
  • Who uses the results of this work, and how often do I talk to them directly?
  • How much of the job is running existing models versus building new methods?
  • Is there support for publishing, conferences or field campaigns?

Related interview guides

Environmental ScientistData Scientist